Derivatives. Partial solutions are ok!

\f\fIndicator function - Indicator function 1{*} l{a true statement} =1 - l{s false statement} = 0 - We can rewrite the loss function of binary logistic regression: m l L(W,b) = i22(1{y'\" = k}logP(y'\" = k I x'\";w,b)) m i=1 k= Loss function of softmax regression . Extend loss function of binary logistic regression to softmax regression m K x() +b( k) L(W,b) = --XX(10y" = k}log- OW (k) m i=l k=1 j=1 where K = P()" = klx";W,b) M j=1